KR20220109364A - Method and apparatus for establishing image recognition model, device, and storage medium - Google Patents

Abstract

Provided are a method and an apparatus for establishing an image recognition model, a device, and a storage medium. The present invention relates to the field of artificial intelligence technology, particularly computer vision and deep learning technology, and can be applied to scenes such as face recognition. The method includes the steps of: obtaining a set of input images; obtaining a trained super-resolution model and a recognition model by jointly training an initial super-resolution model and an initial recognition model using the input image set; and obtaining an image recognition model by dependently combining the trained super-resolution model and the recognition model. A method for establishing an image recognition model according to the present disclosure improves the robustness of the image recognition model against low-quality image data.

Description

METHOD AND APPARATUS FOR ESTABLISHING IMAGE RECOGNITION MODEL, DEVICE, AND STORAGE MEDIUM

The present disclosure relates to the field of artificial intelligence technology, specifically, computer vision and deep learning technology, and in particular, to a method, apparatus, device and storage medium for building an image recognition model, and to be applied to scenes such as face recognition. can

Facial recognition is one of the most widely implemented technologies at the earliest in computer vision technology, and is particularly widely used in security and mobile payment fields. With the widespread application of deep learning in face recognition technology, the accuracy of deep learning-based face recognition has been greatly improved.

However, in the more general unconstrained natural scene, after the camera collects the video stream, the captured facial image is often ambiguous, or the quality degradation such as small face area occurs, resulting in low recognition pass rate or false recognition rate. this is high

The present disclosure provides a method, an apparatus, an apparatus, and a storage medium for building an image recognition model.

According to a first aspect of the present disclosure, there is provided a method of constructing an image recognition model, comprising the steps of: obtaining an input image set; Obtaining a model and a recognition model, and obtaining an image recognition model by subordinately combining the trained super-resolution model and the recognition model.

According to a second aspect of the present disclosure, there is provided an image recognition method, comprising: obtaining an image to be recognized; and adding an image to be recognized to an image recognition model obtained by a method according to any implementation method of the first aspect and outputting a recognition result corresponding to an image to be recognized.

According to a third aspect of the present disclosure, there is provided an apparatus for building an image recognition model, a first acquisition module configured to acquire an input image set, and an initial super-resolution model and an initial recognition model by using the input image set for joint training and a training module configured to obtain a trained super-resolution model and a recognition model, and a combination module configured to obtain an image recognition model by dependently combining the trained super-resolution model and the recognition model.

According to a fourth aspect of the present disclosure, there is provided an image recognition apparatus, comprising: a second obtaining module configured to obtain an image to be recognized; and an image recognition model obtained by a method according to any implementation manner of the first aspect and an output module configured to input an image to be recognized and output a recognition result corresponding to the image to be recognized.

According to a fifth aspect of the present disclosure, there is provided an electronic device, comprising at least one processor and a memory in communication connection with the at least one processor, the memory storing instructions executable by the at least one processor, When the instruction is executed by the at least one processor, it causes the at least one processor to perform the method according to any implementation of the first aspect or the second aspect.

According to a sixth aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium having stored thereon computer instructions for executing the method according to any implementation of the first or second aspect in a computer.

According to a seventh aspect of the present disclosure, there is provided a computer program that, when executed by a processor, implements the method according to any implementation manner of the first aspect or the second aspect.

It is to be understood that the content described herein is not intended to represent key features or critical features of embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will be readily understood from the following description of the specification.

The accompanying drawings are for a better understanding of the present disclosure and do not limit the present disclosure.
1 is an exemplary system architecture applicable to the present disclosure.
2 is a flowchart of one embodiment of a method for building an image recognition model according to the present disclosure.
3 is a schematic diagram of one application scene of the method for building an image recognition model according to the present disclosure.
4 is a flowchart of another embodiment of a method for building an image recognition model according to the present disclosure.
5 is a flowchart of another embodiment of a method for building an image recognition model according to the present disclosure.
6 is a flowchart of one embodiment of an image recognition method according to the present disclosure.
7 is a structural diagram of an embodiment of an apparatus for building an image recognition model according to the present disclosure.
8 is a structural diagram of an embodiment of an image recognition apparatus according to the present disclosure.
9 is a block diagram of an electronic device for implementing a method of constructing an image recognition model according to an embodiment of the present disclosure.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, exemplary embodiments of the present disclosure will be described in conjunction with the drawings, and various details of the exemplary embodiments of the present disclosure are included to aid understanding, and these should be regarded as exemplary only. Therefore, it should be understood by those skilled in the art that various changes and modifications may be made to the embodiments described in the present disclosure without departing from the scope and spirit of the present disclosure. Equally, descriptions of well-known functions and structures are omitted in the following description for clarity and brevity.

In addition, it is intended to explain that the embodiments of the present disclosure and features in the embodiments can be combined with each other as long as there is no contradiction. Hereinafter, the present disclosure will be described in detail in combination with embodiments with reference to the accompanying drawings.

1 shows an exemplary system architecture 100 of an embodiment to which the method for building an image recognition model or the apparatus for building an image recognition model of the present disclosure can be applied.

As shown in FIG. 1 , the system architecture 100 may include terminal devices 101 , 102 , 103 , a network 104 , and a server 105 . The network 104 is used as a medium that provides a communication link between the terminal devices 101 , 102 , 103 and the server 105 . Network 104 may include a variety of connection types, such as wired, wireless communication links, or fiber optic cables.

The user may transmit and receive messages and the like by interacting with the server 105 through the network 104 using the terminal devices 101 , 102 , 103 . Various client applications may be installed in the terminal devices 101 , 102 , and 103 .

The terminal devices 101 , 102 , 103 may be hardware or software. When the terminal devices 101 , 102 , and 103 are hardware, they may be various electronic devices, including, but not limited to, a smart phone, a tablet computer, a laptop computer, and a desktop computer. When the terminal devices 101 , 102 , 103 are software, they may be installed in the electronic device. It may be implemented in a plurality of software or software modules, and may also be implemented in a single software or software module. There is no particular limitation here.

The server 105 may provide various services. For example, the server 105 may analyze and process an input image set obtained from the terminal devices 101 , 102 , and 103 , and generate a processing result (eg, an image recognition model).

The server 105 may be hardware or software. When the server 105 is hardware, it may be implemented as a distributed server cluster consisting of a plurality of servers, or may be implemented as a single server. When the server 105 is software, it may be implemented as a plurality of software or software modules (eg, for providing distributed services), or may be implemented as a single software or software module. There is no particular limitation here.

The method for building the image recognition model provided in the embodiment of the present disclosure is generally executed by the server 105 , and accordingly, the apparatus for building the image recognition model is generally installed in the server 105 .

It should be understood that the number of terminal devices, networks, and servers in FIG. 1 is merely exemplary. Any number of terminal devices, networks and servers may be provided according to the needs of the implementation.

With continued reference to FIG. 2 , a process 200 of one embodiment of a method of building an image recognition model in accordance with the present disclosure is shown. The method of constructing the corresponding image recognition model includes the following steps.

Step 201 obtains an input image set.

In the present embodiment, the executing entity (eg, the server 105 shown in FIG. 1 ) of the method for building an image recognition model may acquire an input image set that may include at least one input image.

The input image in the input image set may be a plurality of images including the face of a person collected in advance through various methods. For example, the input image set may be a plurality of images acquired from an existing image library, and as another example, the input image set is real-time by an image sensor (eg, a camera sensor) in an actual application scene. Since it may be a plurality of collected images, the present disclosure is not particularly limited thereto.

In step 202, the initial super-resolution model and the initial recognition model are jointly trained using the input image set to obtain a trained super-resolution model and a recognition model.

In this embodiment, the executing entity may obtain a trained super-resolution model and a recognition model by jointly training the initial super-resolution model and the initial recognition model using the input image set obtained in step 201 .

Here, the initial super-resolution model and the initial recognition model can be determined in advance, for example, the initial super-resolution model is SRCNN (Super-Resolution Convolutional Neural Networks), FSRCNN (Fast Super-Resolution Convolutional Neural Networks), SRGAN (Super -Resolution Generative Adversarial Network), etc., and the initial recognition model may be a classification recognition model such as the existing ResNet (Residual Network) series, or a model designed according to actual needs.

The execution entity may link training the initial super-resolution model and the initial recognition model using the input image set obtained in step 201, and adjust the parameters of the initial super-resolution model and the initial recognition model through the input image set, , when the joint training stop condition is satisfied, training is stopped to obtain a trained super-resolution model and a recognition model. Among them, the chain training stop condition includes a preset number of training, the value of the loss function no longer decreases, or sets a certain accuracy threshold and stops training when the preset threshold is reached. have.

Step 203 obtains an image recognition model by dependently combining the trained super-resolution model and the recognition model.

In this embodiment, the executing entity may obtain an image recognition model by subordinately combining the trained super-resolution model and the recognition model obtained in step 202 . In this step, by installing the trained super-resolution model before the recognition model, more information can be added to the recognition model, and a better effect can be obtained.

The method for building an image recognition model provided by an embodiment of the present disclosure is first to obtain an input image set, and then, using the input image set, the initial super-resolution model and the initial recognition model are jointly trained to train the second A resolution model and a recognition model are obtained, and finally an image recognition model is obtained by dependently combining the trained super-resolution model and the recognition model. The image recognition model construction method of this embodiment alleviates the influence of images of different resolutions on the classification task by linking training the initial super-resolution model and the initial recognition model, and the robustness of the image recognition model on low-quality data is improved. The recognition accuracy of the recognition model was improved.

In the technical means of the present disclosure, the acquisition, storage, and application of relevant user personal information conforms to the provisions of relevant laws and regulations and does not violate public order and morals.

Continuing to refer to FIG. 3 , FIG. 3 illustrates one application scene of the method for building an image recognition model according to the present disclosure. In the application scene of FIG. 3 , first, the execution entity 301 acquires an input image set 302 . Then, the execution entity 301 jointly trains the initial super-resolution model and the initial recognition model using the input image set 302 to obtain a trained super-resolution model 303 and a recognition model 304 . Finally, the execution entity 301 obtains the image recognition model 305 by subordinately combining the trained super-resolution model 303 and the recognition model 304 .

With continued reference to FIG. 4 , FIG. 4 illustrates a process 400 of another embodiment of a method of building an image recognition model in accordance with the present disclosure. The method of constructing the corresponding image recognition model includes the following steps.

Step 401 obtains an input image set.

Step 401 is basically the same as step 201 of the above embodiment, and for a specific implementation manner, reference may be made to the description of step 201, and a detailed description thereof will be omitted here.

Step 402 calculates the loss function of the initial super-resolution model using the input image set and the reconstructed image set corresponding to the input image set, and updates the parameters of the initial super-resolution model using gradient descent to Get the resolution model.

In the present embodiment, the executing entity (eg, the server 105 shown in FIG. 1 ) of the method for building an image recognition model acquires an input image set, and then restores the image corresponding to each image in the input image set. By determining the image, it is possible to obtain a set of reconstructed images corresponding to the set of input images.

Then, the execution entity calculates the loss function of the initial super-resolution model using the input image in the input image set and the corresponding reconstructed image in the reconstructed image set, and iteratively solves it step by step using gradient descent to minimize the loss You can get function and model parameter values.

Finally, by updating the parameters of the initial super-resolution model with the model parameter values obtained, the quality of the results is improved by obtaining a trained super-resolution model.

Step 403 calculates a loss function of the initial recognition model based on the distances between the image features in the input image set and the reconstructed image set, and updates the parameters of the initial recognition model using gradient descent to obtain the trained recognition model. get

In this embodiment, the execution entity may calculate the loss function of the initial recognition model based on the distance between the image features in the input image set and the reconstructed image set. For example, we can merge the images in the input image set and the reconstructed image set to obtain the final image set, then calculate the distances between image features in the acquired image set and compute the loss function of the initial recognition model based on these distances. have.

Then, using gradient descent, iteratively solves one step at a time to obtain a minimized loss function and model parameter values, and then updates the parameters of the initial recognition model with the obtained model parameter values to obtain a trained recognition model. to improve the classification accuracy of

In some optional implementation manners of this embodiment, the gradient descent method is a stochastic gradient descent method. By using the stochastic gradient descent method, the minimized loss function and model parameter values can be obtained more quickly, increasing the efficiency of model training.

Step 404 connects the output end of the part before the loss function of the trained super-resolution model to the input end of the recognition model to obtain an image recognition model.

In this embodiment, the execution subject may acquire an image recognition model by connecting the output end of the part before the loss function of the trained super-resolution model to the input end of the recognition model. By installing the trained super-resolution model before the recognition model, more information can be added to the recognition model, resulting in better effects.

As can be seen from Fig. 4, compared to the corresponding embodiment of Fig. 2, the method of building the image recognition model of this embodiment emphasizes the step of training the initial super-resolution model and the initial recognition model using the input image set, The efficiency of model training is improved and the accuracy of the trained super-resolution model and recognition model is improved, so that it has a wider application range.

With continued reference to FIG. 5 , FIG. 5 illustrates a process 500 of another embodiment of a method of building an image recognition model in accordance with the present disclosure. The method of constructing the corresponding image recognition model includes the following steps.

Step 501 obtains an input image set.

Step 501 is basically the same as step 401 of the above embodiment, and the detailed implementation method may refer to the description of step 401, so a detailed description thereof will be omitted here.

Step 502 downsamples the images in the input image set to obtain a downsampling image set.

In this embodiment, the executing entity of the method for building an image recognition model (eg, the server 105 shown in FIG. 1 ) performs downsampling of each image in the input image set to obtain a corresponding downsampling image by , may acquire a downsampling image set including a downsampling image corresponding to each input image in the input image set. The downsampling image obtained through this step is a low-quality image that more closely matches the actual application scene.

Step 503 reconstructs images in the downsampling image set using the initial super-resolution model to obtain a reconstructed image set.

In this embodiment, the executing entity reconstructs each downsampling image in the downsampling image set by using the initial super-resolution model to obtain a corresponding reconstructed image, and the reconstructed image is the low-quality image obtained in step 502 . is a high-quality image obtained by restoring

Step 504 calculates the reconstruction loss of the initial super-resolution model based on the input image set and the reconstructed image set, and updates the parameters of the initial super-resolution model using gradient descent to obtain a trained super-resolution model. .

In this embodiment, the execution entity calculates the reconstruction loss using the input image in the input image set and the restored image corresponding to the input image in the restored image set, and iteratively solves it step by step using gradient descent. Thus, the minimized loss function and model parameter values can be obtained. Then, the parameters of the initial super-resolution model are updated with the obtained model parameter values to obtain a trained super-resolution model.

Through the above steps, the result quality of the super-resolution model is improved.

Step 505 merges the input image set, the downsampling image set, and the reconstructed image set to obtain a target image set.

In this embodiment, the execution entity may acquire the target image set by merging the input image set, the downsampling image set, and the reconstructed image set.

Step 506 extracts features of images in the target image set and calculates distances between features of images in the target image set.

In this embodiment, the execution entity may extract a feature of each image in the target image set, and calculate a distance between images in the target image set based on the extracted feature.

Optionally, before acquiring the input image set, the input image in the input image set is annotated and each target object is given an identity document (ID). If the corresponding target object is an object represented by a face in the input image, the input image corresponding to each target object in the input image set will have the same ID, and the ID of the downsampling image and the restored image corresponds to the ID of the input image .

On this basis, in this step, the distance between images can be calculated based on ID, the distance between all images with the same ID can be calculated based on the extracted image features, and then the distance between images with different IDs can be calculated.

Step 507 computes the binary group loss function of the initial recognition model based on the distance, and updates the parameters of the initial recognition model using gradient descent to obtain a trained recognition model.

In this embodiment, the execution entity may calculate the binary group loss function of the initial recognition model based on the distance calculated in step 506 .

Optionally, when two images have the same ID, the loss function at this time is the square of the distance between the two images. When the IDs of two images are different, the loss value at this time is obtained by calculating the margin between the two images and then calculating the max value. That is, since the distance between images with the same ID is closer and the distance between all images with different IDs is longer, the difference between classes increases and the difference within a class decreases.

Then, using the gradient descent method, iteratively solves one step at a time to obtain the minimized loss function and model parameter values, and then uses the obtained model parameter values to update the parameters of the initial recognition model to obtain a trained recognition model.

The classification accuracy of the recognition model is improved through the above steps.

In step 508, an image recognition model is obtained by connecting the output terminal of the part before the loss function among the trained super-resolution models to the input terminal of the recognition model.

Step 508 is basically the same as step 404 of the above embodiment, and detailed implementation manner may refer to the description of step 404, and detailed description thereof will be omitted herein.

As can be seen from FIG. 5 , compared to the embodiment corresponding to FIG. 4 , the method of building the image recognition model of this embodiment is based on the input image set and the reconstructed image set, the reconstruction loss of the initial super-resolution model and the initial recognition By calculating the binary group loss function of the model, and using gradient descent to update the parameters of the initial super-resolution model and the initial recognition model to obtain a trained super-resolution model and recognition model, Classification accuracy was improved.

With continued reference to FIG. 6 , FIG. 6 illustrates a process 600 of one embodiment of an image recognition method in accordance with the present disclosure. The image recognition method includes the following steps.

In step 601, an image to be recognized is acquired.

In this embodiment, the executing entity of the image recognition method (eg, the server 105 shown in FIG. 1 ) may acquire an image to be recognized, where the image to be recognized is an actual application scene of face recognition It may be an image including a human face collected by a camera sensor.

In step 602, an image to be recognized is input to the image recognition model, and a recognition result corresponding to the image to be recognized is output.

In this embodiment, the execution subject may input an image to be recognized into an image recognition model and output a recognition result corresponding to the image to be recognized, wherein the image recognition model is the image recognition model of the above-described embodiment. It may be obtained by the construction method of

After the execution subject inputs the image to be recognized into the image recognition model, the image recognition model first restores the image to be recognized to obtain a corresponding restored image, then extracts the image to be recognized and features of the restored image, and By classifying according to a characteristic, a corresponding recognition result is obtained and the corresponding recognition result is output.

As an image recognition method provided by an embodiment of the present disclosure, first, an image to be recognized is acquired, an image to be recognized is input to an image recognition model, and a recognition result corresponding to the image to be recognized is output. The image recognition method of the present embodiment may improve the accuracy of the recognition result by recognizing an image to be recognized using a previously trained image recognition model.

Continuing to refer to FIG. 7 , as an implementation of the method shown in each figure, the present disclosure provides an embodiment of an apparatus for building an image recognition model, the embodiment of the apparatus is an embodiment of the method shown in FIG. 2 , and the device may be applied to various electronic devices.

As shown in FIG. 7 , the image recognition model building apparatus 700 of this embodiment includes a first acquisition module 701 , a training module 702 , and a combination module 703 . The first acquiring module 701 is configured to acquire an input image set, and the training module 702 is a super-resolution model and a recognition model trained by jointly training the initial super-resolution model and the initial recognition model using the input image set. , and the combining module 703 is configured to subordinately combine the trained super-resolution model and the recognition model to obtain an image recognition model.

In the present embodiment, in the apparatus 700 for constructing an image recognition model, the specific processing of the first acquisition module 701, the training module 702, and the combination module 703 and the technical effects thereof are described in FIG. 2, respectively. Reference is made to the relevant descriptions of steps 201 to 203 of the corresponding embodiment, which are not repeated herein.

In some optional implementations of this embodiment, the training module calculates the loss function of the initial super-resolution model using the input image set and the reconstructed image set corresponding to the input image set, and uses gradient descent to a first update submodule, configured to update a parameter of the resolution model, and calculate a loss function of the initial recognition model based on the distance between the features of the image in the input image set and the reconstructed image set, and use gradient descent to calculate the initial recognition and a second update sub-module, configured to update a parameter of the model.

In some optional implementation manners of this embodiment, the first update submodule includes a downsampling unit, configured to downsample an image in the input image set to obtain a downsampling image set, and downsampling using the initial super-resolution model a reconstruction unit, configured to reconstruct images in the image set to obtain a reconstructed image set; and a first calculation unit, configured to calculate a reconstruction loss of the initial super-resolution model based on the input image set and the reconstructed image set.

In some optional implementation manners of this embodiment, the second update submodule includes: a merging unit, configured to merge the input image set, the downsampling image set, and the reconstructed image set to obtain the target image set; an extraction unit configured to extract features of , a second calculation unit configured to calculate a distance between features of an image in the target image set, and a third calculation unit configured to calculate a binary group loss function of the initial recognition model based on the distance A calculation unit is provided.

In some optional implementation manners of this embodiment, the combining module includes a connecting sub-module, configured to connect an output end of the part before the loss function in the trained super-resolution model to an input end of the recognition model.

Continuing to refer to FIG. 8 , as an implementation of the method shown in each figure, the present disclosure provides an embodiment of an image recognition apparatus, the embodiment of the apparatus corresponding to the embodiment of the method illustrated in FIG. 6 , , the device can be applied to various electronic devices.

As shown in FIG. 8 , the image recognition apparatus 800 of this embodiment includes a second acquisition module 801 and an output module 802 . The second acquisition module 801 is configured to acquire an image to be recognized, and the output module 802 inputs an image to be recognized into the image recognition model and outputs a recognition result corresponding to the image to be recognized.

In the present embodiment, in the image recognition apparatus 800, the specific processing of the second acquisition module 801 and the output module 802 and the technical effects resulting therefrom, respectively, are from steps 601 to steps 601 to the corresponding embodiment of FIG. 6, respectively. Reference is made to the related description of (602), which is not repeated here.

According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

9 shows a schematic block diagram of an exemplary electronic device 900 for implementing an embodiment of the present disclosure. Electronic device refers to various types of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, large computers, and other suitable computers. Electronic device may also refer to various types of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components, their connections and relationships, and their functions presented in the specification are illustrative only and are not intended to limit the implementation of the present disclosure described and/or required herein.

As shown in FIG. 9 , the device 900 includes a computation unit 901 , which is a computer program stored in a read-only memory (ROM) 902 or a random access memory (RAM) in a storage unit 908 . Various appropriate operations and processing can be performed according to the computer program loaded into 903 . In the RAM 903, various programs and data necessary for operating the device 900 may be stored. The calculation unit 901 , the ROM 902 , and the RAM 903 are connected to each other by a bus 904 . An input/output (I/O) interface 905 is also coupled to the bus 904 .

A plurality of members of the device 900 are connected to the I/O interface 905 , for example, an input unit 906 such as a keyboard, mouse, etc., an output unit such as various types of displays, speakers, etc. 907), for example, a storage unit 908, such as a magnetic disk, an optical disk, and a communication unit 909, such as, for example, a network interface card, a modem, a wireless communication transceiver, or the like. The communication unit 909 allows the device 900 to exchange information/data with other devices via a computer network such as the Internet and/or various telecommunication networks.

The computational unit 901 may be a variety of general-purpose and/or dedicated processing assemblies having processing and computational capabilities. Some examples of the computation unit 901 include a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computation units running machine learning model algorithms, and a digital signal processor (DSP). ) and all suitable processors, controllers, microcontrollers, and the like. The calculation unit 901 performs each of the methods and processing described above, such as, for example, a method of constructing an image recognition model or an image recognition method. For example, in some embodiments, a method of constructing an image recognition model or an image recognition method may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 908 . . In some embodiments, some or all of the computer program may be loaded or installed in the device 900 by the ROM 902 and/or the communication unit 909 . When the computer program is loaded into the RAM 903 and executed by the calculation unit 901, one or a plurality of steps of the above-described image recognition model building method or image recognition method can be performed. Alternatively, in another embodiment, the computation unit 901 may be configured to perform the image recognition method or the building method of the image recognition model via any other suitable manner (eg, by firmware).

Various embodiments of the above-described systems and techniques herein include digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), standard product specific (ASSP), system-on-a-chip (SOC), It may be implemented in a complex programmable logic element (CPLD), computer hardware, firmware, software, and/or a combination thereof. These various embodiments may include: That is, implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted in a programmable system comprising at least one programmable processor, the programmable processor being dedicated or A general purpose programmable processor capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and capable of sending data and instructions to a storage system, at least one input device, and at least one output device. can be sent to the device.

Program code for implementing the method of the present disclosure may be written by applying any combination of one or a plurality of programming languages. Such program code is provided to a processor or controller of a general-purpose computer, dedicated computer, or other programmable data processing device, and when the program code is executed by the processor or controller, the functions/operations specified in the flowcharts and/or block diagrams are performed. make it possible The program code may run entirely on the machine or partly on the machine, as a standalone software package, partly on the machine and partly on the remote machine, or completely on the remote machine or server.

In the description of the present disclosure, a machine-readable medium may be a tangible medium, which includes a program used by or in combination with an instruction execution system, device or device, or can be saved The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium includes, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared or semiconductor system, apparatus or apparatus, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include one or more wire-based electrical connections, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory ( EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage, magnetic storage, or any suitable combination of the foregoing.

In order to provide interaction with the user, the system and technology described herein can be implemented in a computer, and the computer is a display device that displays information to the user (for example, a CRT (cathode ray tube) or LCD (liquid crystal display) device) a monitor), a keyboard and a pointing device (eg, a mouse or a trackball), and a user can provide input to the computer by means of the corresponding keyboard and the corresponding pointing device. Other types of devices may also be used to provide interaction with a user, for example, the feedback provided to the user may include any form of sensory feedback (eg, visual feedback, auditory feedback or tactile feedback). ), and may receive a user's input in any format (sound input, voice input, or tactile input).

The systems and techniques described herein can be applied to a computing system including a background member (eg, as a data server), or a computing system including a middleware member (eg, as an application server), or computing including a front end member. a system (eg, a user computer having a graphical user interface or network browser through which a user may interact with embodiments of the systems and technologies described herein), or such a background It may be implemented in a computing system including any combination of members, middleware members, or front end members. Digital data communication in any form or medium (eg, a communication network) may interconnect the members of the system. Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

A computer system may include a client and a server. A client and server are typically remote from each other and typically interact through a communication network. A client-server relationship is created by running a computer program with a client-server relationship on the corresponding computer. The server may be a cloud server, a server in a distributed system, or a server combined with a blockchain.

It should be understood that the various types of processes described above may be used to rearrange, increment, or delete steps. For example, as long as the expected results of the technical solutions disclosed in the present disclosure can be implemented, each step described in the present disclosure may be performed in parallel, sequentially, or in a different order. does not limit this.

The specific embodiments described above do not limit the protection scope of the present disclosure. Those skilled in the art will recognize that various modifications, combinations, subcombinations and substitutions are possible in accordance with design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present disclosure should all fall within the protection scope of the present disclosure.

Claims (16)

acquiring an input image set;
obtaining a trained super-resolution model and a recognition model by jointly training an initial super-resolution model and an initial recognition model using the input image set;
Method of constructing an image recognition model comprising the step of obtaining an image recognition model by dependently combining the trained super-resolution model and the recognition model. The method of claim 1,
The step of jointly training the initial super-resolution model and the initial recognition model using the input image set includes:
calculating a loss function of the initial super-resolution model using the input image set and the reconstructed image set corresponding to the input image set, and updating parameters of the initial super-resolution model using gradient descent;
calculating a loss function of an initial recognition model based on distances between the input image set and image features in the reconstructed image set, and updating parameters of the initial recognition model using gradient descent. 3. The method of claim 2,
Calculating the loss function of the initial super-resolution model by using the input image set and the reconstructed image set corresponding to the input image set,
downsampling images in the input image set to obtain a downsampling image set;
obtaining a restored image set by reconstructing an image in the downsampling image set using the initial super-resolution model;
and calculating a reconstruction loss of the initial super-resolution model based on the set of input images and the set of reconstructed images. 4. The method of claim 3,
Calculating the loss function of the initial recognition model based on the distance between the input image set and the image feature in the reconstructed image set,
merging the input image set, the downsampling image set and the reconstructed image set to obtain a target image set;
extracting features of images in the target image set;
calculating distances between features of images in the set of target images;
calculating a binary group loss function of the initial recognition model based on the distance. 3. The method of claim 2,
The gradient descent method is a stochastic gradient descent method. The method of claim 1,
The step of subordinately combining the trained super-resolution model and the recognition model,
and connecting the output end of the part before the loss function of the trained super-resolution model to the input end of the recognition model. obtaining the image to be recognized; and
The step of inputting the image to be recognized into an image recognition model obtained by the method of constructing an image recognition model according to any one of claims 1 to 6, and outputting a recognition result corresponding to the image to be recognized An image recognition method comprising a. a first acquiring module configured to acquire an input image set;
a training module configured to jointly train an initial super-resolution model and an initial recognition model using the input image set to obtain a trained super-resolution model and a recognition model;
and a combination module configured to obtain an image recognition model by dependently combining the trained super-resolution model and the recognition model. 9. The method of claim 8.
The training module is
a first update, configured to calculate a loss function of an initial super-resolution model by using the input image set and a reconstructed image set corresponding to the input image set, and update parameters of the initial super-resolution model by using gradient descent sub-modules,
a second update submodule, configured to calculate a loss function of an initial recognition model based on a distance between the input image set and a feature of an image in the reconstructed image set, and update the parameters of the initial recognition model by using gradient descent A device comprising: 10. The method of claim 9.
The first update sub-module,
a downsampling unit, configured to downsample images in the input image set to obtain a downsampling image set;
a reconstructing unit, configured to reconstruct an image in the downsampling image set by using the initial super-resolution model to obtain a reconstructed image set;
a first calculation unit, configured to calculate a reconstruction loss of the initial super-resolution model based on the input image set and the reconstructed image set. 11. The method of claim 10.
The second update sub-module,
a merging unit, configured to merge the input image set, the downsampling image set, and the reconstructed image set to obtain a target image set;
an extraction unit configured to extract features of images in the target image set;
a second calculation unit, configured to calculate a distance between features of an image in the target image set;
and a third calculation unit, configured to calculate a binary group loss function of the initial recognition model based on the distance. 9. The method of claim 8.
The combination module is
and a connection submodule, configured to connect an output end of the part before the loss function of the trained super-resolution model to an input end of the recognition model. a second acquisition module configured to acquire an image to be recognized;
A configuration to input the image to be recognized into an image recognition model obtained by the method of constructing an image recognition model according to any one of claims 1 to 6, and output a recognition result corresponding to the image to be recognized An image recognition device comprising an output module being at least one processor;
and a memory in communication connection with the at least one processor;
An instruction executable by the at least one processor is stored in the memory, and when the instruction is executed by the at least one processor, the at least one processor performs the method of any one of claims 1 to 7 An electronic device to perform. A non-transitory computer-readable storage medium having stored thereon computer instructions for executing the method of any one of claims 1 to 7 in a computer. A computer program that, when executed by a processor, implements the method of any one of claims 1-7.

Images